Examples may include a storage appliance having a mass storage device and a compute engine communicating peer-to-peer with each other, with the compute engine including a programmable logic component to execute a function to read data from the at least one storage device, process the data, and write data to the at least one storage device.
|
1. A storage appliance comprising:
a bus;
at least one mass storage device coupled to the bus;
a library of functions, wherein each function is received from at least one of a cloud service provider operating a data center and a customer of the data center; and
at least one compute engine coupled to the bus to communicate peer-to-peer with the at least one mass storage device, the at least one compute engine including a programmable logic component to execute at least one function to read data from the at least one mass storage device, process the data; and write data to the at least one mass storage device, wherein the programmable logic component comprises at least two configurable slots to store a compute offload bitstream including first and second functions, a first configurable slot to store the first function and a second configurable slot to store the second function, wherein the first function is to send output data to the second function.
5. A system comprising:
a network;
at least one compute server coupled to the network, the at least one compute server to execute an application; and
at least one storage appliance coupled to the network, storage appliance comprising:
a bus;
at least one mass storage device coupled to the bus;
a library of functions, wherein each function is received from at least one of a cloud service provider operating a data center and a customer of the data center; and
at least one compute engine coupled to the bus to communicate peer-to-peer with the at least one mass storage device, the at least one compute engine including a programmable logic component to execute at least one function to read data from the at least one mass storage device, process the data; and write data to the at least one mass storage device, wherein the programmable logic component comprises at least two configurable slots to store a compute offload bitstream including first and second functions, a first configurable slot to store the first function and a second configurable slot to store the second function, wherein the first function is to send output data to the second function.
9. A data center system comprising:
a network; and
a plurality of server racks coupled to the network, wherein each server rack includes
a plurality of compute servers to execute applications; and
a plurality of storage appliances, wherein each storage appliance includes
a bus;
a plurality of mass storage devices coupled to the bus;
a library of functions, wherein each function is received from at least one of a cloud service provider operating a data center and a customer of the data center; and
a plurality of compute engines coupled to the bus to communicate peer-to-peer with the plurality of mass storage devices, each compute engine including a programmable logic component to execute at least one function to read data from the mass storage devices, process the data; and write data to the mass storage devices, wherein the programmable logic component comprises at least two configurable slots to store a compute offload bitstream including first and second functions, a first configurable slot to store the first function and a second configurable slot to store the second function, wherein the first function is to send output data to the second function.
2. The storage appliance of
3. The storage appliance of
4. The storage appliance of
6. The system of
7. The system of
8. The system of
10. The data center system of
11. The data center system of
12. The data center system of
|
Examples described herein are generally related to accelerating processing of functions in a data center.
Function as a Service (FaaS) is a fast-growing usage model for data centers. FaaS is a category of cloud computing services that provides a computing platform to customers to allow customers to develop, run, and manage application functions without the complexity of building and maintaining the infrastructure typically associated with developing and launching an application. Building an application following this model is one way of achieving a “serverless” architecture and is typically used when building microservices applications.
FaaS enables data centers (e.g., such as those run by cloud service providers) to provide to their customers the capability to select specific functions to execute in the data center and to pay for per function processing. Well-defined functions (e.g., sort, search, matrix operations, etc.) are provided by cloud service providers to customers. In addition, customers can provide pre-programmed modules (e.g., as plugins, virtual machines (VMs) or containers) that are run as functions on data center servers. When these custom functions are both compute intensive and data intensive with small data results, as is typical in applications such as analytics, artificial intelligence (AI), and others, processing of custom functions by the data center becomes inefficient. This is because the data is typically read by the custom functions as they are being executed on the data center compute server CPUs, where the data is processed, and the results of the computations are sent back to the customer's system and/or to storage devices within the data center. Processing by compute server CPUs, data transfers over the network, and heavy usage of memory bandwidth all frequently become bottlenecks to various degrees in such use cases, thus slowing down overall application throughput and increasing response times. Additionally, when customers are paying per processing core usage or per function execution, the typical data center architecture becomes too expensive.
In some scenarios, field programmable gate array (FPGAs) are used to execute the functions. However, the FPGAs are resident in the compute servers. The compute servers control the data flow to and from the FPGAs vis-à-vis mass storage devices in the data center, execute additional processing, and transfer results of the FPGA computations back to the requesting system component. This still results in heavy network and memory bandwidth usage, negatively affecting application performance.
As contemplated in the present disclosure, embodiments of the present invention comprise a data center architecture that allows FaaS customers to specify storage-centric compute offloads for functions. As used herein, a function is a sequence of instructions that perform a specific task, packaged as a unit. This unit can then be used wherever that particular task should be performed. Compute offloads are structured as a combination of zero or more cloud service provider functions or zero or more customer provided functions and are executed by a storage appliance instead of a compute server. Further, data center customers can package and execute functions in such a way that the functions can be chained together. Embodiments enable compute offloads to any compute engine in the data center for cost and power reduction (not just those resident within compute servers), while providing flexible software and hardware accelerators desired by cloud service customers. In embodiments, the compute engines are included close to mass storage devices in the data center for efficient data transfers and computations, such as within a storage appliance. The present architecture results in significant reduction in data traffic between storage devices and compute servers, thereby improving system performance for cloud applications.
According to some examples, compute server 101, as shown in
In some examples, compute server 101, includes, but is not limited to, a server, a server array or server farm, a web server, a network server, an Internet server, a work station, a mini-computer, a main frame computer, a supercomputer, a network appliance, a web appliance, a distributed computing system, multiprocessor systems, processor-based systems, or a combination thereof. Also, circuitry 120 having processing cores 122-1 to 122-m may include various commercially available processors, including without limitation Intel® Atom®, Celeron®, Core (2) Duo®, Core i3, Core i5, Core i7, Itanium®, Pentium®, Xeon® or Xeon Phi® processors; and similar processors. Circuitry 120 may include at least one cache 135 to store data.
According to some examples, primary memory 130 and/or NVM 165 is composed of one or more memory devices or dies which may include various types of volatile and/or non-volatile memory. Volatile types of memory include, but are not limited to, dynamic random-access memory (DRAM), static random-access memory (SRAM), thyristor RAM (TRAM) or zero-capacitor RAM (ZRAM). Non-volatile types of memory include byte or block addressable types of non-volatile memory having a 3-dimensional (3-D) cross-point memory structure that includes chalcogenide phase change material (e.g., chalcogenide glass) hereinafter referred to as “3-D cross-point memory” (such as 3D XPoint™ commercially available from Intel Corporation). Non-volatile types of memory also include other types of byte or block addressable non-volatile memory such as, but not limited to, multi-threshold level NAND flash memory, NOR flash memory, single or multi-level phase change memory (PCM), resistive memory, nanowire memory, ferroelectric transistor random access memory (FeTRAM), magneto-resistive random-access memory (MRAM) that incorporates memristor technology, spin transfer torque MRAM (STT-MRAM), or a combination of any of the above. In one example, primary memory 130 is a volatile memory and persistent memory 165 is an NVM.
In an embodiment, each compute server of
In an embodiment, each server rack includes one or more storage appliances, such as storage appliance 1 232, storage appliance 2 234, . . . storage appliance N 236, where N is a natural number. In practice, each server rack may include tens, hundreds, or even thousands of storage appliances. Storage appliances communicate with other storage appliances and compute servers over network 211. Storage appliances also communicate with client systems over external network 210. In embodiments of the present invention, a storage appliance comprises a system to provide both mass storage and computing capabilities. In mass storage devices currently used in data centers, the mass storage devices typically are relatively “dumb” storage devices and do not provide significant compute capabilities. In contrast, in embodiments of the present invention the data center offloads at least some of the compute load of the compute servers in the data center to the storage appliances.
Data center 212 provides FaaS capabilities to data center customers operating client systems. One or more modules called functions, such as data center function (DCFN) 1 214, DCFN 2 216, . . . DCFN Q 218, where Q is a natural number, are available in the data center for use in processing the compute workloads of customers. Each DCFN, when executed on a computer server or a storage appliance, performs a well-defined operation based on zero or more input data and resulting in one or more output data. For example, data center 212 provides functions to sort data, search data, and compute matrix operations. Many other functions are contemplated and embodiments of the present invention are not limited to these examples. Similarly, client system 202 defines customer provided functions, such as customer function CFN 1 204, CFN 2 206, . . . CFN R 208, where R is a natural number. Customer provided functions may, when executed, perform any processing needed as part of an application. In an embodiment, data center 212 charges a fee to a client system based on what functions (whether data center provided or customer provided) are executed by components in the server racks. The fee may be based on one or more of many different possible variables and statistics of the data center, such as the number of times a function was executed, how many processor cycles were used, how much storage was used, how much network bandwidth was used, and so on. A client system selects zero or more data center provided functions and zero or more customer provided functions to be executed as part of an application. In an embodiment, input data for a function is obtained from a client system and/or one or more storage appliances, and output data is written to a client system and/or one or more storage appliances.
When some functions are executed by compute servers, data center 212 may experience negative application performance effects from having to obtain data to process from one or more storage appliances, communicate the data over network 211, process the data in the compute servers (including accessing memories such as primary memory 130 and non-volatile memory 165 by storing data obtained from storage appliances into these memories before, during, and after processing the data), and communicating resulting data back over network 211 to storage appliances. To improve system performance, in embodiments of the present invention data center 212 offloads at least some of the compute workload from compute servers to storage appliances.
In an embodiment, storage appliance 400 includes a function library 416 to store one or more functions (zero or more DCFNs and zero or more CFNs). In an embodiment, function library 416 is stored in one or more of the storage devices and is read by any compute engine for processing. By storing functions in function library 416, the functions are accessible within the storage appliance. Thus, once a function is stored for the first time within the storage appliance, the function need not be retrieved from a compute server, another storage appliance, or a client system (unless and until the function is changed). This locality of the functions within the storage appliance results in better overall system performance.
In various embodiments, a compute engine resident in the storage appliance may be a programmable logic component comprising a FPGA, a group of FPGAs, a central processing unit (CPU), a graphics processing unit (GPU), an application specific integrated circuit (ASIC), or any other programmable logic component. In various embodiments, the plurality of compute engines within a storage appliance comprises any combination of zero or more FPGAs, zero or more CPUs, and zero or more GPUs, with each compute engine including at least one programmable logic component (i.e., a FPGA, CPU, GPU, or ASIC). In an embodiment, a programmable logic component such as a CPU, GPU, ASIC, or FPGA in the storage appliance executes a function to read data from a storage device, process the data, write data to the same or another storage device, consume data obtained from network 211 or 210 or another chained programmable logic component, and share data to network 211 or 210 or to another chained programmable logic component.
In an embodiment, a bootstrap bitstream is used to load the FPGA. The bootstrap bitstream is downloaded from a data center server (such as a compute server) during startup of the storage appliance containing the FPGA. The bootstrap bitstream provides the basic functionality of configuring PCIe blocks, memory controller 508, network controller 512, and other components.
FPGA 501 includes a plurality of configurable slots 514, 516, where different compute offloads can be loaded and executed. A configurable slot comprises a set of Logic Elements (LEs) that can be used by a compute offload. A compute offload as used herein includes one or more functions, either provided by the data center, the customer, or both. The location of LEs for each configurable slot may be fixed in the FPGA and a unique bitstream based on one or more functions can be generated for each configurable slot.
In an embodiment, compute offloads have a defined data-plane interface allowing the FPGA to chain compute offloads into pipelines of compute offloads. These pipelined compute offloads can be added or removed from the pipeline using partial reconfiguration of the FPGA. Although an FPGA is shown in
The example of
For example, there may be functions in function library 416 to perform tasks such as decrypting data, decompressing data, encrypting data, compressing data, querying data, averaging data, and so on. Functions can include any processing specified by a data center (e.g., DCFNs) or a customer (e.g., CFNs). In embodiment, such functions are grouped into pipelined compute offloads using a plurality of configurable slots in FPGA 501. In an embodiment, providing the capability to chain compute offloads enables a customer to specify multiple offloads, each having varying M inputs and N output streams, and to be able to construct a data flow graph that connects them. This capability allows functions in some situations to be performed in parallel (thereby improving application performance), to be developed independently, and to be reused in various combinations.
Various components of compute server 101 and storage appliance 304 are communicatively coupled to each other by various types of communications media to coordinate operations. The coordination may involve the uni-directional or bi-directional exchange of information. For instance, the components communicate information in the form of signals communicated over the communications media. The information can be implemented as signals allocated to various signal lines. In such allocations, each message is a signal. Further embodiments, however, may alternatively employ data messages. Such data messages are sent across various connections. Example connections include parallel interfaces, serial interfaces, and bus interfaces.
It should be appreciated that the exemplary computing platforms shown in the Figures may represent one functionally descriptive example of many potential implementations. Accordingly, division, omission or inclusion of block functions depicted in the accompanying figures does not infer that the hardware components, circuits, software and/or elements for implementing these functions would necessarily be divided, omitted, or included in embodiments.
One or more aspects of at least one example may be implemented by representative instructions stored on at least one machine-readable medium which represents various logic within the processor, which when read by a machine, computing device or system causes the machine, computing device or system to fabricate logic to perform the techniques described herein. Such representations, known as “IP cores” may be stored on a tangible, machine readable medium and supplied to various customers or manufacturing facilities to load into the fabrication machines that actually make the logic or processor.
Various examples may be implemented using hardware elements, software elements, or a combination of both. In some examples, hardware elements may include devices, components, processors, microprocessors, circuits, circuit elements (e.g., transistors, resistors, capacitors, inductors, and so forth), integrated circuits, ASIC, programmable logic devices (PLD), digital signal processors (DSP), FPGA, memory units, logic gates, registers, semiconductor device, chips, microchips, chip sets, and so forth. In some examples, software elements may include software components, programs, applications, computer programs, application programs, system programs, machine programs, operating system software, middleware, firmware, software modules, routines, subroutines, functions, methods, procedures, software interfaces, application program interfaces (API), instruction sets, computing code, computer code, code segments, computer code segments, words, values, symbols, or any combination thereof. Determining whether an example is implemented using hardware elements and/or software elements may vary in accordance with any number of factors, such as desired computational rate, power levels, heat tolerances, processing cycle budget, input data rates, output data rates, memory resources, data bus speeds and other design or performance constraints, as desired for a given implementation.
Some examples may include an article of manufacture or at least one computer-readable medium. A computer-readable medium may include a non-transitory storage medium to store logic. In some examples, the non-transitory storage medium may include one or more types of computer-readable storage media capable of storing electronic data, including volatile memory or non-volatile memory, removable or non-removable memory, erasable or non-erasable memory, writeable or re-writeable memory, and so forth. In some examples, the logic may include various software elements, such as software components, programs, applications, computer programs, application programs, system programs, machine programs, operating system software, middleware, firmware, software modules, routines, subroutines, functions, methods, procedures, software interfaces, API, instruction sets, computing code, computer code, code segments, computer code segments, words, values, symbols, or any combination thereof.
Some examples may be described using the expression “in one example” or “an example” along with their derivatives. These terms mean that a particular feature, structure, or characteristic described in connection with the example is included in at least one example. The appearances of the phrase “in one example” in various places in the specification are not necessarily all referring to the same example.
Included herein are logic flows or schemes representative of example methodologies for performing novel aspects of the disclosed architecture. While, for purposes of simplicity of explanation, the one or more methodologies shown herein are shown and described as a series of acts, those skilled in the art will understand and appreciate that the methodologies are not limited by the order of acts. Some acts may, in accordance therewith, occur in a different order and/or concurrently with other acts from that shown and described herein. For example, those skilled in the art will understand and appreciate that a methodology could alternatively be represented as a series of interrelated states or events, such as in a state diagram. Moreover, not all acts illustrated in a methodology may be required for a novel implementation.
A logic flow or scheme may be implemented in software, firmware, and/or hardware. In software and firmware embodiments, a logic flow or scheme may be implemented by computer executable instructions stored on at least one non-transitory computer readable medium or machine readable medium, such as an optical, magnetic or semiconductor storage. The embodiments are not limited in this context.
Some examples may be described using the expression “coupled” and “connected” along with their derivatives. These terms are not necessarily intended as synonyms for each other. For example, descriptions using the terms “connected” and/or “coupled” may indicate that two or more elements are in direct physical or electrical contact with each other. The term “coupled,” however, may also mean that two or more elements are not in direct contact with each other, but yet still co-operate or interact with each other.
It is emphasized that the Abstract of the Disclosure is provided to comply with 37 C.F.R. Section 1.72(b), requiring an abstract that will allow the reader to quickly ascertain the nature of the technical disclosure. It is submitted with the understanding that it will not be used to interpret or limit the scope or meaning of the claims. In addition, in the foregoing Detailed Description, it can be seen that various features are grouped together in a single example for the purpose of streamlining the disclosure. This method of disclosure is not to be interpreted as reflecting an intention that the claimed examples require more features than are expressly recited in each claim. Rather, as the following claims reflect, inventive subject matter lies in less than all features of a single disclosed example. Thus, the following claims are hereby incorporated into the Detailed Description, with each claim standing on its own as a separate example. In the appended claims, the terms “including” and “in which” are used as the plain-English equivalents of the respective terms “comprising” and “wherein,” respectively. Moreover, the terms “first,” “second,” “third,” and so forth, are used merely as labels, and are not intended to impose numerical requirements on their objects.
Although the subject matter has been described in language specific to structural features and/or methodological acts, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are disclosed as example forms of implementing the claims.
Trika, Sanjeev N., Khan, Jawad B., Wysocki, Piotr
Patent | Priority | Assignee | Title |
Patent | Priority | Assignee | Title |
9038189, | Feb 24 2011 | STEVENS, ANDREW G ; JANSEN, MATTHEW J; EGNOR, KYLE A | System and method for programming a trading system |
20020020175, | |||
20040016244, | |||
20040024495, | |||
20140143121, | |||
20180337904, |
Executed on | Assignor | Assignee | Conveyance | Frame | Reel | Doc |
Jun 01 2018 | TRIKA, SANJEEV N | Intel Corporation | ASSIGNMENT OF ASSIGNORS INTEREST SEE DOCUMENT FOR DETAILS | 046235 | /0144 | |
Jun 04 2018 | KHAN, JAWAD B | Intel Corporation | ASSIGNMENT OF ASSIGNORS INTEREST SEE DOCUMENT FOR DETAILS | 046235 | /0144 | |
Jun 06 2018 | Intel Corporation | (assignment on the face of the patent) | / | |||
Jun 06 2018 | WYSOCKI, PIOTR | Intel Corporation | ASSIGNMENT OF ASSIGNORS INTEREST SEE DOCUMENT FOR DETAILS | 046235 | /0144 |
Date | Maintenance Fee Events |
Jun 06 2018 | BIG: Entity status set to Undiscounted (note the period is included in the code). |
Jun 21 2023 | M1551: Payment of Maintenance Fee, 4th Year, Large Entity. |
Date | Maintenance Schedule |
Jan 28 2023 | 4 years fee payment window open |
Jul 28 2023 | 6 months grace period start (w surcharge) |
Jan 28 2024 | patent expiry (for year 4) |
Jan 28 2026 | 2 years to revive unintentionally abandoned end. (for year 4) |
Jan 28 2027 | 8 years fee payment window open |
Jul 28 2027 | 6 months grace period start (w surcharge) |
Jan 28 2028 | patent expiry (for year 8) |
Jan 28 2030 | 2 years to revive unintentionally abandoned end. (for year 8) |
Jan 28 2031 | 12 years fee payment window open |
Jul 28 2031 | 6 months grace period start (w surcharge) |
Jan 28 2032 | patent expiry (for year 12) |
Jan 28 2034 | 2 years to revive unintentionally abandoned end. (for year 12) |